Gentics Mesh Search Plugin

Gentics Mesh Search Plugin

Description

Gentics Mesh already has great search capabilities - but this plugin gets you even further: It provides two ready-to-use endpoints that allows your frontend to use the search and autocomplete functionality without having to build your own Elasticsearch query.

This plugin will accept a simple string as a query, and will perform the transformation to a full-featured elasticsearch query, taking care of languages, tag filtering and project scopes - so you don’t have to deal with that in your frontend. If the user’s query only returns few matching nodes, you can configure to receive "did you mean" results in addition.

You can furthermore configure which fields you want to return for each search result. If you want, you can even extend the given functionality with custom Java code!

Documentation

A plugin to search Mesh projects used as a content respository for the Gentics CMS.

Installation

Make sure to only use commercial plugins which match the Major and Minor version of the Gentics Mesh server. Plugins which do not match may not be compatible with the Gentics Mesh version.

Commercial plugins can be downloaded from our maven site. Alternatively you can also use maven to download the jar:

mvn dependency:get \
  -Dartifact=com.gentics.mesh.plugin.commercial:mesh-search-plugin:$YOUR_MESH_VERSION \
  -DremoteRepositories=gtx-commercial::default::https://maven.gentics.com/maven2-commercial \
  -Ddest=mesh-search-plugin.jar -Dtransitive=false

If you get an "Unauthorized" error, please locate your maven settings (usually found in ~/.m2/settings.xml) and add our server to the servers list:

settings.xml
<settings>
 ...
 <servers>
    <server>
      <id>gtx-commercial</id>
      <username> $YOUR_USER_ID </username>
      <password> $YOUR_API_KEY </password>
    </server>
 ...

Once downloaded, place the jar file, optionally together with a config file and other assets, in the configured plugins folder of your Mesh installation — then the plugin(s) will automatically be deployed during server startup.

Configuration

This is an example configuration file specifying every possible setting. Except for projects and customQueries (which are by default empty), all values are the default values.

Example configuration:
pageSize: 8
projects:
  main_project_name:
    - other_searched_project
resultQuery: null
didYouMeanThreshold: 10
customQueries:
  - custom_search
  - name: custom_auto
    type: AUTOCOMPLETE
caching: true
expireTime: 300

Note that whatever is specified by projects each Mesh project will always search itself. This setting is only useful to define additional projects that should be searched.

For custom queries of type SEARCH a shorthand notation can be used by just specifying the name (without the name: and type: parts).

Search endpoints

When no custom handlers or queries are configured, the plugin exposes two search endpoints. One for normal search, and one for autocompletion.

Table 1. Normal search summary

Endpoint

/api/v1/[project]/plugins/search/search

Handler

com.gentics.mesh.plugin.handler.SearchHandler

Used queries

search.es, didYouMeanBinary.es, didYouMeanContent.es, searchResults.graphql

Parameters

Name Description Default

q

Query String

REQUIRED

p

Current page

0

t

Filter by tags (facetted search)

[]

l

Language

Portal default language

branch

Project branch

Empty

Request flow

The processing of a new request starts in the SearchHandlers internalHandle() method.

Sections in the request flow wich cover only internals are marked with an asterisk (*).

Request validation

The request will fail when the mandatory parameter q is missing.

Loading the Elasticsearch query

The search query is loaded by the QueryService, which will look for the query in the following places in this order

  1. [pluginStorageDir]/queries/[project]/search.es

  2. [pluginStorageDir]/queries/search.es

  3. classpath:/queries/search.es

(whichever is found first, will be used). Depending on the used QueryBuilder the file must contain valid JSON, or may contain Handlebars placeholders.

The QueryBuilder implementations provided by the plugin will replace the following parameters in the loaded query.

Table 2. Searchquery parameters
Name Description

query

Sanitized search query

lang

Language

projects

Regex, matching all searched project names

tags

Comma separated list from parameter t

Both the included BasicQueryBuilder and the TemplateQueryBuilder will add paging parameters automatically, after the above parameters have been inserted. The size parameter is the configured pageSize and from is computed from the current page (request parameter p) and the pageSize.

By default the TemplateQueryBuilder is used, so a search query might look like the one below (more examples for the two different QueryBuilders can be found in the /examples/queries directory).[1]

Note that the query builder will already sanitize the parameters, so the {{{ parameter }}} notation should be used, to prevent Handlebars from escaping them again.

The TemplateQueryBuilder will cache the created template for expireTime seconds.

Example search.es
{
  "query": {
    "bool": {
      "must": {
        "multi_match": {
          "type": "best_fields",
          "query": "{{{ query }}}",
          "fields": [
            "fields.title.search^3",
            "fields.teaser.search^2",
            "fields.content.search",
            "fields.name.search"
          ]
        }
      },
      "should": {
        "match_phrase": {
          "fields.content.search": "{{{ query }}}"
        }
      },
      "filter": [
        {
          "regexp": {
            "project.name.raw": {
              "value": "{{{ projects }}}",
              "flags": "ANYSTRING"
            }
          }
        },
        {
          "regexp": {
            "language.raw": {
              "value": "{{{ language }}}",
              "flags": "ANYSTRING"
            }
          }
        },
        {
          "match": {
            "fields.searchtags": {
              "query": "{{{ tags }}}",
              "operator": "and",
              "zero_terms_query": "all"
            }
          }
        }
      ]
    }
  },
  "aggregations": {
    "searchtags": {
      "terms": {
        "field": "fields.searchtags.taglist",
        "size": 9999,
        "min_doc_count": 1
      }
    }
  },
  "_source": [
    "project.name",
    "uuid",
    "language",
    "fields.title"
  ]
}
Using the BasicQueryBuilder*

To use the BasicQueryBuilder instead of the TemplateQueryBuilder, the respective binding in the BindModule needs to be changed.

The BasicQueryBuilder will insert the parameters by setting them for certain paths in the raw query. These paths must be specified in a field called params. Any parameters for which no paths are defined, will not be inserted.

Example query for BasicQueryBuilder
{
  "params": {
    "query": [
      "bool/must/multi_match/query",
      "bool/should/match_phrase/fields.content.search"
    ],
    "project": [
      "bool/filter[0]/regexp/project.name.raw"
    ],
    "language": [
      "bool/filter[1]/regexp/language.raw"
    ]
  },
  // ... rest as in example above
}
Performing the search request*

The plugin will send the final Elasticsearch query to

  • /api/v1/[meshProject]/rawSearch/nodes if only the current Mesh project is searched, or to

  • /api/v1/rawSearch/nodes if multiple projects are searched

From the resulting JSON, the search hits, total count, hits count and aggregations are extracted.

Refining the search hits

If a resultQuery is configured, the GraphQl query with that name will be loaded from the QueryService and used to load additional information about the hits from Mesh. All hits will be grouped by Mesh branch and one GraphQl query will be issued per branch.

The resultQuery can only contain the following Handlebars placeholders for schema names, and must otherwise be a valid GraphQL query:

  • {{{ binaryContentSchema }}},

  • {{{ contentSchema }}}, and

  • {{{ folderSchema }}}

The search result query must retrieve the uuid and language fields, and should load the path.

The variable uuids will be set to the hit UUIDs and the language will be passed in the variable lang.

Minimum resultQuery
query ($uuids: [String]) {
  nodes(uuids: $uuids) {
    elements {
      languages {
        uuid
        language
        path
      }
    }
  }
}

The languages part can be extended to load specific fields:

Example resultQuery loading fields
query ($uuids: [String], $lang: String) {
  nodes(uuids: $uuids, lang: [ $lang ]) {
    elements {
      languages {
        uuid
        language
        path
        ... contentFields
      }
      breadcrumb {
        ... breadcrumbFields
      }
    }
  }
}

fragment contentFields on {{{ contentSchema }}} {
  title
  teaser
}

fragment breadcrumbFields on {{{ folderSchema }}} {
  name
  navhidden
}

When a fragment definition cannot be found in the query, the plugin will search for a file with the name of the fragment (and the .graphql extension). Those are loaded with the same priorities mentioned for the search query:

  1. queries/[project]/fragments/fragmentName.graphql,

  2. queries/fragments/fragmentName.graphql, and

  3. classpath:queries/fragments/fragmentName.graphql

This makes it possible to fetch different fields for search hits, based on the project that received the search request.

The response of the resultQuery is used to enhance the hits:

Table 3. Hits enhancement

Name

Description

uuid

Set as uuid

language

Set as language

path

Used to build the url of the hit

Every entry in languages is passed to the injected HitTransformer, which can further enhance the hits with data loaded by the GraphQL query.

Default implementation is com.gentics.mesh.plugin.search.transformer.BasicHitTransformer, which adds the field title as excerpt, and copies all fields to the result.

Cleaning up the search result*

Any null entries are removed by finalizeHits() (which can happen, if the resultQuery did not return data for an entry in the raw hits), and addPagingInfo() repairs info about total hits, if some were removed before and adds link for next page of search results (if more hits are expected).

Did-You-Mean suggestions

If less than didYouMeanThreshold[2] hits are returned, the 'did-you-mean' elasticsearch queries (didYouMeanBinary.es and didYouMeanContent.es) are loaded from the QueryService for each searched project. (Currently, the queries can be specific for the project that receives the search request, but not for every searched project, this should be changed).

The QueryService uses the injected QueryBuilders buildDidYouMeanQuery() to create the final search queries for each project.

For each project, the queries are POSTed to /api/v1/[project]/rawSearch/nodes and from all responses, the option with highest score is added as didYouMean to the complete search result.

Example didYouMeanContent.es
{
  "suggest": {
    "text": "{{{ query }}}"
    "did-you-mean-teaser": {
      "term": {
        "suggest_mode": "always",
        "field":  "fields.teaser.suggest"
      }
    },
    "did-you-mean-content": {
      "term": {
        "suggest_mode": "always",
        "field":  "fields.content.suggest"
      }
    }
  }
}
Example for a search response
{
  "results": [
    {
      "branch": "12345",
      "uuid": "1",
      "language": "de",
      "url": "/foo/bar.html",
      "title": "Testseite",
      "isBinary": false,
      "excerpt": "Testseite"
    },
    {
      "branch": "12345",
      "uuid": "2",
      "language": "de",
      "url": "/doo/tmp.html",
      "title": "Eine weitere Seite",
      "isBinary": false,
      "excerpt": "Eine weitere Seite"
    },
  ],
  "totalHits": 2,
  "tags": [
    {
      "Verschiedenes": 1
    },
    {
      "Tests": 2
    }
  ],
  "hasMore": false,
  "didYouMean": "Taste"
}

Autocomplete

Table 4. Normal search summary

Endpoint

/api/v1/[project]/plugins/search/autocomplete

Handler

com.gentics.mesh.plugin.handler.AutocompleteHandler

Used queries

autocomplete.es

Parameters

Name Description Default

q

Query String

REQUIRED

l

Language

Portal default language

branch

Project branch

Empty

Request flow

The flow for an autocomplete request is very similar to the one for a normal search, so the following sections are not covered with the same level of detail.

The processing of a new request starts in the AutocompleteHandlers internalHandle() method.

Request validation

The request will fail when the mandatory parameter q is missing.

Loading the Elasticsearch query

As with the normal search the autocomplete query is loaded by the QueryService, which will look for the query in the following places in this order

  1. [pluginStorageDir]/queries/[project]/autocomplete.es

  2. [pluginStorageDir]/queries/autocomplete.es

  3. classpath:/queries/autocomplete.es

The QueryBuilder implementations provided by the plugin will replace the following parameters in the loaded query.

Table 5. Searchquery parameters
Name Description

query

Sanitized search query

lang

Language

projects

Regex, matching all searched project names

Requirements for autocomplete queries

The autocomplete query should find the same results as the normal search query, so the actual query part should be the same with the exeption that the autocomplete query should use phrase_prefix wherever applicable. Since autocompletion does not support tags, any parts of the query concerning them can be removed.

The provided QueryBuilders will not add the size parameter to the query, so it should be hardcoded in the file (with a low value since the search results are only needed to generate suggestions).

In contrast do a normal search query, an autocomplete query must define highlights for the searched fields, which are later used to extract suggestions. On the other hand aggregations and _source fields are not used by the handler (although specifying _source with a short list reduces the size of the Elasticsearch result).

For the highlights it is important to define special pre_tags and post_tags, because by default these are <strong> tags, which might occur in the source.

Example search.es
{
  "query": {
    // Like the search query without tags and using "phrase_prefix".
  },
	"highlight": {
		"pre_tags" : [ "%hb%" ],
		"post_tags" : [ "%he%" ],
		"fragment_size": 0,
		"number_of_fragments": 0,
		"fields": {
			"fields.content.autocomplete": {},
			"fields.teaser.autocomplete": {},
			"fields.title.autocomplete": {}
		}
	},
  "_source": [ "uuid" ]
}
Performing the search request*

As for the normal search the plugin will send the final Elasticsearch query to

  • /api/v1/[meshProject]/rawSearch/nodes if only the current Mesh project is searched, or to

  • /api/v1/rawSearch/nodes if multiple projects are searched

Extracting the suggestions

From the resulting JSON, the highlights are searched for suggestions by matching agains the special pre and post tags %hb% and %hw% respectively. Any complete words that are already part of the query term are removed (as well as any HTML markup), and the resulting words are returned as a list of suggestions.

Example for a search response

When requesting autocompletion for the term test the following might be returned by the autocomplete handler:

[
  {
    "name": "tester"
  },
  {
    "name": "testing"
  },
  {
    "name": "testenvironment"
  }
]

Administrative endpoints

The search plugin also registers the following endpoints to control caching of search queries. All the endpoints only listen to POST requests.

  • /api/v1/[project]/plugins/search/cache/clear: clears all caches

  • /api/v1/[project]/plugins/search/cache/enable: clears all caches, and enables caching

  • /api/v1/[project]/plugins/search/cache/disable: clears all caches, and disables caching

Custom endpoints

There are two ways to add more endpoints via the search plugin: specifying a custom search query and implementing the CustomHandler interface.

Custom search query

By specifying a list of customQueries in the plugin configuration. The name of the custom query is used to

  • load the raw query using the name as the filename (without the extions), and

  • register the endpoint /api/v1/[project]/plugins/search/custom/[name].

Depending on the type of the custom query, either the SearchHandler or the AutocompleteHandler will be executed using the specified query instead of search.es or autocomplete.es respectively.

When no type for a custom query is specified, the SearchHandler will be used implicitly.

For example the following configuration will add the endpoints …​/custom/contacts and …​/custom/contactAutocomplete which will use the specified queries for search and autocompletion:

[yaml] ` customQueries: - name: contacts type: SEARCH - name: contactsAutocomplete type: AUTOCOMPLETE `

(Note that the first entry could simply be written as - contacts because the type is then implicitly SEARCH.)

Custom handler

When using a custom raw query is not enough, the CustomHandler interface can be implemented.

Every custom handler must expose its name and type via the respective methods.

For custom handlers with very complex behaviour it is best to override the default handle() method which handles the request completely on its own.

When only specific post processing is needed (which is outside the scope of a HitTransformer) it is also possible to just implement the getQuery() and processResult() methods. In this case the SearchHandler is invoked but it will get the query from the custom handlers getQuery() method and will pass the result to processResult() (which is responsible for terminating the request). The default implementation will just cast the result to a JsonObject and send it as the response.

Custom handler providers

While custom queries only need entries in the plugin configuration, custom handlers need additional Java code, and the implemented handlers must be made available to the plugin.

To do this a CustomHandlerProvider must be implemented and bound in the BindModule.

The default CustomHandlerProvider will return an empty set.

Included queries

Default resultsQuery

A default result query with the name searchResults is provided by the plugin, which will load the fields uuid, language, path and fields.teaser, as well as the breadcrumbs with the fields name and navhidden.

Tipps

Highlighting

Since the indexed fields might contain HTML markup themselves, it is probably a good idea to define custom pre and post tags for Elasticsearch to mark highlights. These custom markers should then be transformed to appropriate markup by a HitTransformer.


1. The queries loaded from the files need not be valid JSON in general but the TemplateQueryBuilder as well as the BasicQueryBuilder will only work on valid JSON queries.
2. To disable did-you-mean suggestions, just set the value to 0.

Plugin Details

The plugin includes the required logic to create the query based on your search terms, and to render the resulting search hits.

Interested?

Version

1.7.0

License

commercial

Authors

Gentics